test_seasonal_order fails on i386

Warn on use and xfail

Author: Rebecca N. Palmer <rebecca_palmer@zoho.com>
Forwarded: no

Gbp-Pq: Name xfail_i386_ets.patch

diff --git a/statsmodels/tsa/exponential_smoothing/ets.py b/statsmodels/tsa/exponential_smoothing/ets.py
index ca354b0576351d70b65f6e512459700e571da3df..bab8aee4341c38d09c68399393469b8c383c3cae 100644 (file)
--- a/statsmodels/tsa/exponential_smoothing/ets.py
+++ b/statsmodels/tsa/exponential_smoothing/ets.py
@@ -171,6 +171,13 @@ from statsmodels.tsa.exponential_smoothing.initialization import (
 )
 from statsmodels.tsa.tsatools import freq_to_period
 
+import warnings
+import sys
+import platform
+import re
+warn_ets = "ETSModel can give wrong results on 32 bit i386" if (bool(re.match('i.?86|x86',platform.uname()[4])) and sys.maxsize<2**33) else False
+
+
 # Implementation details:
 
 # * The smoothing equations are implemented only for models having all
@@ -426,6 +433,8 @@ class ETSModel(base.StateSpaceMLEModel):
         missing="none",
     ):
 
+        if warn_ets:
+            warnings.warn(warn_ets)
         super().__init__(
             endog, exog=None, dates=dates, freq=freq, missing=missing
         )

diff --git a/statsmodels/tsa/tests/test_exponential_smoothing.py b/statsmodels/tsa/tests/test_exponential_smoothing.py
index 642b2aad3592a62a7baa6e21a0eb8e8e3fa7356f..6ef8ff1a6b27e66fdf2ff4d0a29fa25a66a4afae 100644 (file)
--- a/statsmodels/tsa/tests/test_exponential_smoothing.py
+++ b/statsmodels/tsa/tests/test_exponential_smoothing.py
@@ -13,7 +13,7 @@ import pandas as pd
 import pytest
 import scipy.stats
 
-from statsmodels.tsa.exponential_smoothing.ets import ETSModel
+from statsmodels.tsa.exponential_smoothing.ets import ETSModel,warn_ets
 import statsmodels.tsa.holtwinters as holtwinters
 import statsmodels.tsa.statespace.exponential_smoothing as statespace
 
@@ -1052,6 +1052,7 @@ def test_estimated_initialization_short_data(oildata, trend, seasonal, nobs):
 
 
 @pytest.mark.parametrize("method", ["estimated", "heuristic"])
+@pytest.mark.xfail(condition=bool(warn_ets),reason='known to fail on i386',strict=False)
 def test_seasonal_order(reset_randomstate, method):
     seasonal = np.arange(12.0)
     time_series = np.array(list(seasonal) * 100)